| Popis: |
The Arctic coast is facing rapid, irreversible changes mainly caused by Climate Warming, e.g., melting sea ice, permafrost thaw, glacial retreat, land uplift and sea level rise. These processes are leading to fundamental changes in the ecosystem structure and functioning, negatively impacting biological and human communities. Under this complex setting, more knowledge is needed to identify the hotspots of shoreline displacement at an Arctic scale. Thus, the goal of this study was to develop and describe a procedure for mapping long-term shoreline displacement in the Arctic that can provide local communities and environmental managers better opportunities to adapt to further coastal changes. Therefore, the procedure will need to be transferrable to diverse environments and able to handle pan-Arctic analyses at a 30-meter spatial resolution. In this study, the procedure was developed using two test areas: Tanafjorden in the low Arctic mainland Norway and Kongsfjorden in the high Arctic Svalbard. The presentation introduces the final procedure and validation results, and discusses its applicability to pan-Arctic shoreline displacement analyses.The procedure was calibrated in the surroundings of Tanafjorden. It was built on a 40-year time-series of open Landsat and Sentinel multispectral satellite images, taken during the Arctic summer. Supervised random forest classification was used to identify land and water pixels, utilizing information from multiple infrared bands and spectral indices. Mountain shadow pixels were treated as their own class and then merged to the land class. Open spatial data were used for limiting the area-of-interest and for automated creation of training data. In total over 700 individual images were first classified separately to account for local environmental conditions and transient illumination conditions. Images were then summarized over 5-year time-steps. The classification results were examined against an independent validation dataset of 2000 land cover observations and manually digitized shoreline, and the supervised classification results were compared to single-band classifications based on Otsu’s thresholding. The final procedure was then validated in the Kongsfjorden environment. The process was built on Google Earth Engine’s image collections and cloud computing infrastructure to minimize computing times.The results indicate that it is possible to transform open satellite imagery into 40-year pan-Arctic shoreline displacement information, with a 30-meter resolution and an overall accuracy of more than 95 %. Data fusion is needed in most processing steps: to limit the area-of-interest, save computing power and reduce errors, provide information that complements multispectral satellite data and reduce the impact of short-term atmospheric and water-level effects. Summarizing dozens of images efficiently removes data gaps and the impact of noise, but this efficiency is sensitive to the number of summarized images. The single-image classification approach is flexible and seems to make the procedure transferable to different locations. Cloud image collections are needed to remove the bottleneck of reading and writing satellite data, and potentially allows the promising procedure to be applied at a pan-Arctic scale in the future. |
